EP1121754B1 - Schnelle faltung unter verwendung einer ungeraden tranformation - Google Patents

Schnelle faltung unter verwendung einer ungeraden tranformation Download PDF

Info

Publication number
EP1121754B1
EP1121754B1 EP99970801A EP99970801A EP1121754B1 EP 1121754 B1 EP1121754 B1 EP 1121754B1 EP 99970801 A EP99970801 A EP 99970801A EP 99970801 A EP99970801 A EP 99970801A EP 1121754 B1 EP1121754 B1 EP 1121754B1
Authority
EP
European Patent Office
Prior art keywords
overlap
blocks
frequency
overlaps
length
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Lifetime
Application number
EP99970801A
Other languages
English (en)
French (fr)
Other versions
EP1121754A1 (de
Inventor
Richard Hellberg
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Telefonaktiebolaget LM Ericsson AB
Original Assignee
Telefonaktiebolaget LM Ericsson AB
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Telefonaktiebolaget LM Ericsson AB filed Critical Telefonaktiebolaget LM Ericsson AB
Publication of EP1121754A1 publication Critical patent/EP1121754A1/de
Application granted granted Critical
Publication of EP1121754B1 publication Critical patent/EP1121754B1/de
Anticipated expiration legal-status Critical
Expired - Lifetime legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H03—ELECTRONIC CIRCUITRY
    • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
    • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
    • H03M13/03—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words
    • H03M13/23—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using convolutional codes, e.g. unit memory codes
    • H—ELECTRICITY
    • H03—ELECTRONIC CIRCUITRY
    • H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
    • H03H17/00—Networks using digital techniques
    • H03H17/02—Frequency selective networks
    • H03H17/0211—Frequency selective networks using specific transformation algorithms, e.g. WALSH functions, Fermat transforms, Mersenne transforms, polynomial transforms, Hilbert transforms
    • H03H17/0213—Frequency domain filters using Fourier transforms

Definitions

  • the present invention relates generally to the problem of filtering, decimation or interpolation and frequency conversion in the digital domain, and more particularly to the use of a modified fast convolution algorithm in wideband multichannel receiver, channelization, and transmitter, de-channelization, structures of a radio communication system.
  • Channelization and de-channelization can be defined as the filtering, decimation/interpolation and the frequency conversion of the signals transmitted and received.
  • the traditional receiver architecture as seen in FIG. 1 can be explained in terms of the Radio Frequency (RF) signal being received by the antenna 105 and then downconverted to an intermediate frequency (IF) by an RF front end 110.
  • the RF front end 110 consists of components such as Low Noise Amplifiers (LNA's), filters and frequency conversion circuits.
  • LNA's Low Noise Amplifiers
  • the desired channel is then extracted by the receiver channelizer 120.
  • the channelizer 120 also consists of LNA's, frequency conversion circuits and filters.
  • the desired channel is then processed at baseband by the RX baseband processing unit 130 to produce the received digital data stream.
  • Today baseband processing usually consists of analog-to-digital conversion, digital filtering, decimation, equalization, demodulation, channel decoding, de-interleaving, data decoding, timing extraction etc.
  • the traditional transmitter architecture in FIG. 1. is the dual of the receiver architecture.
  • the transmitted data is first processed by the TX baseband processing unit 140 which consists of data coding, interleaving, channel coding, modulation, interpolation filtering, digital-to-analog conversion etc.
  • the baseband channel is then converted to an IF frequency via the transmit de-channelizer 150.
  • the transmit de-channelizer 150 consists of filters, frequency conversion circuits and low power amplifiers.
  • the IF signal is then converted to RF and amplified by the RF front end 160 which consists of frequency conversion circuits, filters, and a high power amplifier.
  • the signal is then transmitted by the antenna 165.
  • Figure 1 illustrates the traditional architecture for a single channel receiver and transmitter as used in a mobile terminal (i.e. mobile phone) application.
  • a basestation multiple channels are processed in a similar way.
  • the path will split at some point to form multiple paths for each channel being processed.
  • the channels will be processed individual and then they will be combined at some point to form a multichannel signal.
  • the point of the split and combination varies, and therefore a variety of basestation receiver and transmitter architectures can be created. More importantly, though, the traditional analog and digital interface is currently somewhere between the channelizer and baseband processing blocks.
  • the analog channelizer/dechannelizer is complex to design and manufacture, and therefore costly. Therefore, in order to provide a cheaper and more easily produced channelizer/de-channelizer, the future analog and digital interface will lie, instead, somewhere between the RF front end and channelizer blocks. Future radio receiver and transmitter structures of this type are called a variety of names, including multistandard radio, wideband digital tuners, or wideband radio and software defined radio, and they all require a digital channelizer/de-channelizer.
  • Efficient digital channelizer/de-channelizer structures consisting of filtering, decimation/interpolation and frequency conversion, are very important in terms of power consumption and die area on a per channel basis.
  • One of the main goals being to integrate as many channels into a single Integrated Circuit (IC) as possible there are several known ways to achieve digital channelization/dechannelization.
  • IC Integrated Circuit
  • ADC Integrated Circuit
  • the wideband signal is centered at an Intermediate Frequency (IF) and typically consists of many Frequency Division Multiplexed (FDM) channels.
  • IF Intermediate Frequency
  • FDM Frequency Division Multiplexed
  • This receiver architecture mimics the functions of a traditional analog channelizer with In-phase and Quadrature(IQ) frequency conversion using e.g. sin/cos generators, decimating and filtering on a per-channel basis.
  • IQ In-phase and Quadrature
  • the bulk of the decimation filtering can be done with computationally cheap CIC filters.
  • Integrated circuits containing this architecture are readily available from several manufacturers. The dual of this architecture is also possible for the transmitter.
  • the IQ channelizer is flexible in that it can handle many standards simultaneously and that the channels can be placed arbitrarily. Its main drawback is the need for an IQ frequency conversion at a high input sampling frequency and subsequent decimation filters for each channel. This means that the die area and power consumption is relatively high per channel.
  • FIG. 3 Another channelizer possibility is to build a decimated filter bank in the receiver, as shown in Figure 3. This method shares a common polyphase filter between many, or all, channels. The hardware cost for this structure is small since it is split between many channels, and good filtering can be achieved. Filter banks are also good for use in transmitter de-channelizers since they both interpolate and add the channels together. An example of this is illustrated in WO 9528045 "Wideband FFT Channelizer".
  • decimated filter bank has a very low cost per channel, but only if all or the majority of channels are used. This architecture is also very inflexible since the channels have to lie on a fixed frequency grid and only one channel spacing is possible. Multiple standards make the filter bank concept require multiple sampling rates, which means multiple architectures, including the ADC and channelizer, are required for simultaneous multiple standards.
  • a variation on the structure of the decimated filter bank, called a subsampled filter bank, can lower the computational cost at the expense of flexibility. For example, requirements for adaptive channel allocation, irregular channel arrangements and frequency hopping precludes using subsampled filter banks, since all channels must be available at the same time.
  • the third main channelization technique is based on the fast convolution scheme of the overlap-add (OLA) or overlap-save (OLS) type.
  • Fast convolution is a means of using cyclic convolution to exactly perform linear convolution, i.e. Finite Impulse Response (FIR) filtering.
  • FIR Finite Impulse Response
  • a state of the art fast convolution algorithm is shown conceptually in Figure 4.
  • the input data is divided into overlapping blocks in the Block Generator. These blocks are discrete Fourier-transformed in the DFT (Discrete Fourier Transform) and subsequently multiplied point-by-point with a filter response in the frequency domain. This filter response can be obtained by discrete Fourier-transforming the impulse response of a filter.
  • the blocks are then transformed back to the discrete time domain by the Inverse DFT (IDFT) and added together in the Block Combiner.
  • IDFT Inverse DFT
  • the stand-alone modified fast convolution algorithm in the prior art performs all the filtering alone, without any additional signal processing. This method leads to various delays.
  • delays are an inherent part of satellite systems, due to the time to transmit to and from the satellite.
  • delays due to the filtering method effect the system proportionately less than if the stand-alone modified fast convolution algorithm were to be used in a radio, e. g. cellular, system. In most radio systems the delay becomes a much more crucial factor which should be reduced as much as possible.
  • the stand-alone modified fast convolution algorithm applied to the receiver channelizer, chops the incoming data signal into blocks whose size depends on the percentage of overlap (%overlap) and the length of the Discrete Fourier Transform (DFT).
  • the DFT is subsequently performed.
  • the truncated filter response that is the number of filter coefficients (N coefficients ) is less than the length of the DFT (N DFT ), is implemented directly in the frequency domain. This is accomplished by multiplying the filter coefficients with the selected output bins of the DFT.
  • IDFT Inverse Discrete Fourier transform
  • the blocks are then overlapped, depending on the %overlap, and combined.
  • the combination is either a process of adding the overlapped section, overlap and add, or discarding the overlapped section, overlap and save. Note that overlap/add and overlap/save can be considered two extremes, and there are techniques known in the art that lie in-between these two.
  • a frequency domain implementation for replacing convolution by multiplication of transforms is described by E.R. Ferrara, "Frequency Domain Implementations of Periodically Time-Varying Filters", IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 33, no. 4, pages 883-892 .
  • the truncation of the frequency response in the stand-alone modified fast convolution algorithm distinguishes it from the standard fast convolution approach. It causes the circular convolution algorithm to now only approximate linear convolution, although with carefully chosen coefficients the error can be kept small. Truncation of the frequency response also performs decimation by a factor of (N coefficients /N DFT ), and the frequency conversion is completed by centering the truncated filter coefficients on the wanted channel.
  • the truncated frequency response also causes a dramatic reduction in the computational complexity in the channel specific parts of the algorithm, that is everything but the DFT.
  • the number of multiplications needed to implement the frequency filter and the size of the IDFT are reduced by approximately a factor of (N coefficients /N DFT ).
  • the stand-alone modified fast convolution algorithm can also be applied to the transmitter de-channelizer, containing all the same attributes.
  • DFT discrete cosine transform
  • FFT Fast Fourier Transform
  • two real data blocks can be processed at the same time in one complex DFT processor. Some extra adders and memory are then needed for post-processing. This is more efficient than using two dedicated real DFTs.
  • Pruning refers to the process of cutting away branches in the DFT that do not affect the output. The output points that are not needed are never computed.
  • a computational reduction can also be achieved if the complex multiplication of the filter frequency response is replaced by real multiplication and a subsequent circular shift of the IDFT output block of data before it is combined to form the time domain samples of the desired channel.
  • the amount of circular shift depends only on the %overlap and the length of the IDFT.
  • the improved modified fast convolution algorithm as described in WO99/65172 improves upon the stand-alone modified fast convolution when applied to radio communication systems, see Figure 2. It splits the necessary filtering between the MFC algorithm and additional channel filtering, thereby improving the power consumption, die area and computational complexity when compared to the prior art stand-alone MFC. It is also a very flexible algorithm in terms of designing it for combination of different systems parameters, sampling frequency, channel bandwidth, channel separation and bit-rate.
  • a further advantage in WO99/65172 is that the MFC part of the algorithm processes smaller-sized blocks and therefore produces smaller delays, delays which become acceptable for land-base radio communication systems.
  • the improved M FC algorithm is considered to be very suitable for e. g. channelization/de-channelization in a wide variety of radio communication systems. It is therefore a good choice for future hardware platforms that will support multiple standards for more than a few channels activated at any one time.
  • the present invention in this application is applicable to both the stand-alone MFC as found in"A Flexible On-Board Demultiplexer/Demodulator", discussed above, and the improved MFC as found in WO99/65172 .
  • MFC modified fast convolution
  • the number of points in the DFT and IDFT are powers of two. Since the lengths of the input and output transforms are both powers of two, overlaps of 50%, 75&, 25% (generally k*1/2 n ) are possible.
  • the decimation and interpolation ratios are limited to powers of two (N FFT /N IFFT ).
  • the successive blocks are, in the state of the art algorithm, multiplied in the frequency domain with identical filter responses, H(k).
  • the decimation ratio is 1024/32, i.e. 32.
  • the overlap can be 50%, 75%, 25%, down to k*1/32, where "k" is an integer.
  • the IFFT size is instead chosen to be 64, the decimation ratio is 16 and the smallest granularity is considered to one bin, therefore the granularity of the overlap is 1/64 of the block length (possible overlaps k*1/64).
  • the present invention relates generally to the problem of filtering, decimation or interpolation and frequency conversion in the digital domain, to the use of a modified fast convolution algorithm in wideband multichannel receiver, channelization, and transmitter, de-channelization, structures of a radio communication system, and more particularly to the problems discussed above.
  • the means of solving these problems according to the present invention are summarized in the following.
  • the modified fast convolution algorithm as used in the e. g. channelization/de-channelization structures of WO99/65172 is limited because the input and output transforms are both powers of two. This limits the number of possibilities in the decimation and interpolation ratios for channelization and dechannelization, respectively. Because these transform sizes are limited to powers of two, other parameters in the algorithm are then fixed to values achievable with this set of transform sizes.
  • transform sizes in the state of the art are powers of 2, it is not obvious to see that it is possible to use transform sizes in either end that also has other factors in them than two, like three or five. This works, without substantial modifications to the algorithm, as long as both the input and output transform have factors in common to provide for the overlap. For example, a common factor of four makes possible 50%, 75% and 25% overlap, or a common factor of three would make possible overlaps of 33% and 66%.
  • the present invention makes it possible to use input and output transforms in the modified fast convolution algorithm whose sizes contain factors other than powers of two, e.g. three or five. This works, without substantial modifications to the algorithm, as long as both the input and output transform have factors in common to provide for the overlap. For example, a common factor of four makes possible 50%, 75% and 25% overlap, or a common factor of three would make possible overlaps of 33% and 66%. In the preferred embodiment, other modifications will be added to the invention which will extend the use of the modified fast convolution algorithm to input and output transforms which have no common factors.
  • Figure 6 shows the situations in which the present invention are used.
  • the MFC is divided into situations where the input and output transforms either have a common factor (CF) or do not have a common factor (No CF) in common with the overlap.
  • CF common factor
  • No CF common factor
  • On the left side of the chart are shown three situations for the sizes of the transforms. In the first one both the input and output transform sizes are a power of two. In the second one, one of the transforms is a power of two while the other is a power of two further multiplied by some other integer. In the third example both transform sizes can be any integer n 1 and n 2 , where both of them may either have or not have a common factor with the other.
  • the invention can be described as consisting of essentially 2 steps.
  • the first step is to make sure that we use different overlaps on consecutive blocks that, on average, give the same overlap on both the input and output ends.
  • the second step is to align the signal in consecutive blocks of time.
  • a third step of an advantageous embodiment is then to compensate for phase shifts due to frequency shifting.
  • the essence of the invention is that it decouples the input and output transform lengths in the modified fast convolution algorithm from each other and from the overlap. it makes it possible to use any transform length on the input together with any transform length on the output and at the same time use any overlap. This provides an enormous amount of freedom compared with the limitations of state of the art. Input and output sample rates can now be chosen with much finer resolution, and decimation and interpolation ratios need no longer be powers of two.
  • the present invention is not discussed in terms of any particular system. It is particularly applicable to many radio base station applications in e.g. cellular, Land Mobile Network (LMR), satellite, wireless local area networks (WLAN's). However, it is not limited to these systems and may, in general, be used in any system using the modified fast convolution algorithm. In addition, it's use is not restricted to use in basestations, but may also be used in e.g. future mobile terminals that are also capable of handling multiple channels simultaneously.
  • LMR Land Mobile Network
  • WLAN's wireless local area networks
  • the method according to the present invention separates into three varieties: (1) letting the input transform determine the overlap, (2) letting the output transform determine the overlap, or (3) choosing an overlap that is independent of both the input and output block lengths.
  • These solutions contain the same ingredients, as the inventive aspect is the same, but will look slightly different in their specific implementations. Examples from the first two varieties will be included in the following description of the solution steps.
  • the first step 710 is making sure that we use different overlaps on consecutive blocks that, on average, give the same overlap as on the opposite end of the algorithm.
  • the second step 720 is to align the signal in consecutive blocks of time.
  • a third step 730 of a advantageous embodiment is then to compensate for phase shifts due to frequency shifting.
  • the overlap is denoted l/m, then in general m different overlaps will have to be used on one or both ends of the algorithm. If l and m have a common factor, they can both first be reduced by this factor. We can then create vectors of length m, representing the lengths of either the overlapping or the non-overlapping parts of the blocks, that on average give the overlap l/m .
  • the sequence of overlaps (or non-overlaps) will repeat cyclically, although it is also possible to implement the invention so that the overlaps come randomly, as long as the average on both sides of the algorithm is the same. If m is a factor in the length of one of the transforms, that end can have the same overlap for all blocks.
  • the first step 710 can be illustrated with an example where we assume that the input transform has even length, the output transform has odd length and the overlap is 50% on the input.
  • the overlap on the output could then be separated into two overlaps, one of odd length and one of even length, that are used interleavingly on every other block .
  • the input transform is 128 points and the output transform is 25 points.
  • the second step 720 of the invention we align the signal in the consecutive blocks in time.
  • the blocks' starting points do not come regularly, as would occur when the overlaps are the same.
  • the time between the first sample of the blocks on the output end could be 13 and 14 samples, respectively. This yields an average of 13.5.
  • the time alignment is done by time-shifting the signal within the block so that it compensates for the slight misalignment of the starting time of the respective blocks. This can be done by multiplying the DFT samples coming from different blocks by sinusoids with different incremental phase shifts (the DFTs of the different delays). An equivalent, simpler and less computationally complex approach would be to multiply the coefficients of the filter response in the frequency domain, H(k), with the same incremental phase shift. This means that a set of m filter responses, corresponding to the m different time shifts, is needed.
  • the time alignment is calculated via the difference between the relative starting points for the blocks on the input and output ends of the algorithm, which is the same as the accumulated difference between the input and output non-overlapping parts.
  • Figure 9 illustrates this concept.
  • n 1 and n 2 are the lengths of the input and output transforms
  • n olp1 (q) and n olp2 (q) see 1190, 1195, respectively, Figure 10
  • the time compensation factor for block one, Tc(1) can often be set to zero, but a certain time shift can also be added equally to all blocks (by setting Tc(1) to a value other than zero) in order to minimize the maximum absolute time shift.
  • a third step 730 of a method according to an advantageous embodiment of the present invention is to compensate for phase shifts due to frequency shifting.
  • Figure 10 illustrates two shifts, Shift #1 and Shift #2, possible in the modified fast convolution algorithm as implemented in a channelizer/de-channelizer.
  • a frequency shift, #1 is included which is performed by using a certain range of the frequency domain samples coming from the DFT 1020 in a channelizer 1000 or by inserting the filtered DFT samples at a certain place in the large IDFT of a dechannelizer.
  • the channelizer 1000 this can be viewed as if the lowest selected bin of the DFT 1020 is shifted down to zero frequency, and in the dechannelizer as if the zero bin of the DFT is shifted up to the lowest frequency of the channel.
  • a shift of the bins in the frequency domain corresponds to a multiplication of the time samples in a block of size n by a sinusoid, e j2pi*f/n*t , where f is the frequency shift (an integer) and t is the number of a sample in the block of size n.
  • f the frequency shift (an integer)
  • t the number of a sample in the block of size n.
  • the phase compensation is done by calculating to which phase the modulating sinusoid has moved during the non-overlapping part of the block and shift the phase of the next block accordingly, by multiplying the whole block with a constant phasor. After a certain number of blocks, in general the same as the number of different timeshifts, m, the phase has returned to its initial value.
  • phase compensation can also be incorporated into the set of filter responses by multiplying the elements of each filter response with a constant phasor, since the number of different phase shifts that are needed generally is the same as the number of different time shifts.
  • phase compensation also depends on the frequency shift, which means that an individual set of filter responses is generally required for each channel in the channelizer.
  • n is the length of the transform
  • n olp (q) is the length of the overlapping part of the q-th block
  • f shift is the frequency shift.
  • the phase compensation for the first block, Pc(1) can be set to zero.
  • the values of accumulated non-overlapping parts times the frequency shift that are above n can be reduced modulo n, since this number represents one full circle of the phasor.
  • the values of accumulated non-overlapping parts themselves can of course also be reduced modulo n.
  • phase compensation is the same for both Shift #1 and Shift #2, but the lengths of the transforms are in general different and the accumulated non-overlapping parts are in general also different and have to be calculated separately.
  • the shift associated with inserting the filtered DFT samples at a certain place in the large IDFT is a positive shift, which must be remembered when using the formula for phase compensation.
  • the accumulated value from the previous blocks is 77,154,230 and 307 for blocks 2 through 5.
  • the phase compensation would then become 0, 2pi*77/128*f Shift1 , 2pi*154/128*f Shift1 , 2pi*230/128*f Shift1 and 2pi*307/128*f Shift1 . Since 307+77, 384, is divisible by 128 the phase will return to zero after five blocks, and the sequence of phase compensations can be repeated.
  • phase compensation due to the first shift is the same as for systems having transforms with common factors and uniform overlaps. This would also be the case for other systems where the overlap is determined by the DFT size in the case of a receiver (channelizer) or determined by the IDFT size in the case of a transmitter (dechannelizer).
  • This phase compensation is quite simple to implement since it is calculated modulo 2 in the case of 50% overlap and modulo 4 in the case of 75% and 25% overlaps. It is also computationally cheap since the multiplication of the blocks by the two or four uniformly spaced phasors are just multiplications by plus and minus one or multiplications by plus and minus one and swapping the real and imaginary parts of the signal.
  • This phase compensation has previously been thoroughly described in U. S. Patent publication No. US2001025291, titled "Flexibility Enhancement to the Modified Fast Convolution Algorithm" filed on September 18, 1998 to Leyonhjelm et al.
  • phase compensation due to a cyclic shift within the smaller range would be dependent on the different overlaps on this end.
  • this compensation would be 0 and 2pi*13/27*f Shift2 on consecutive blocks.
  • the present invention can be divided into three varieties: (1) letting the input transform determine the overlap, (2) letting the output transform determine the overlap, or (3) choosing an overlap that is independent of both the input and output block lengths.
  • n 1 *(m-l)/m equals 146 ⁇ 2 7 we let the input vector of non-overlapping parts be [146 147 146 146 147 146] which, as one of many possibilities, averages 146 ⁇ 2 7 . In the same way n 2 *(m-l)/m equals 13 ⁇ 1 7 so the output non-overlap vector [13 13 13 14 13 13 13] is chosen.
  • Tc(p) The vector of incremental phase shifts, Tc(p), corresponding to the time alignments of blocks 1 through 7 becomes (all numerators modulo 256 and 23, respectively)
  • f shift1 is a negative
  • the fully implemented system is shown conceptually in Figure 11.
  • the figure illustrates that there are several frequency responses 1130 that each has a time alignment dependent on the input and output overlap and transform lengths.
  • the phase compensations 1170, 1175 depend on the overlap, transform length and shift on each end and are also performed on each of the frequency responses 1130.
  • the preferred implementation as described above uses m different frequency responses 1130 into which all time alignments 1180 and phase compensations 1170, 1175, are absorbed. This means that these frequency responses 1130 can be computed once and then used for a certain channel as long as desired, which implies a low computational cost at the expense of increased memory needed for storing these different responses 1130.
  • time alignments 1180 and phase compensations 1170, 1175 can be multiplied to the blocks in real time, minimizing storage.

Landscapes

  • Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Pure & Applied Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Computing Systems (AREA)
  • Algebra (AREA)
  • Computer Hardware Design (AREA)
  • Probability & Statistics with Applications (AREA)
  • Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)
  • Complex Calculations (AREA)
  • Transmission Systems Not Characterized By The Medium Used For Transmission (AREA)

Claims (10)

  1. Verfahren zum Erhöhen der Anzahl nutzbarer Transformationslängen und Überlappungen in einem Signal, das in einem Algorithmus für eine modifizierte schnelle Faltung mit Eingabelänge n1 und Ausgabelänge n2 und Überlappung l/m transformiert ist, und wobei es nicht der Fall ist, dass n1 und n2 beide gleich einer Potenz von 2 sind, wobei das Verfahren umfasst:
    Bereitstellen von Überlappungen auf aufeinanderfolgenden Blöcken auf einer Seite, wobei die Blöcke nicht die selbe Länge haben, und die im Durchschnitt die selbe Überlappung wie die andere Seite ergeben; und
    Ausrichten des Signals in aufeinanderfolgenden Blöcken in der Zeit.
  2. Verfahren zum Erhöhen der Anzahl nutzbarer Transformationslängen und Überlappungen in einem Signal, das in einem Algorithmus für eine modifizierte schnelle Faltung mit Eingabelänge n1 und Ausgabelänge n2 und Überlappung l/m transformiert ist, wobei das Verfahren umfasst:
    Bereitstellen von Überlappungen auf aufeinanderfolgenden Blöcken auf einer Seite, die im Durchschnitt die selbe Überlappung wie die andere Seite ergeben; und
    Ausrichten des Signals in aufeinanderfolgenden Blöcken in der Zeit,
    wobei n1 und n2 beide gleich einer Potenz von 2 sind, und der Überlappungsnenner, m, keinen gemeinsamen Faktor mit n1 oder n2 hat.
  3. Verfahren gemäß Anspruch 1 oder 2, ferner dadurch gekennzeichnet, dass:
    die Blöcke frequenzverschoben werden; und
    Phasenverschiebungen aufgrund der Frequenzverschiebung kompensiert werden.
  4. Verfahren gemäß Anspruch 1, ferner dadurch gekennzeichnet, dass:
    der Überlappungsnenner, m, keinen gemeinsamen Faktor mit n1 oder n2 hat.
  5. Verfahren gemäß Anspruch 1, ferner dadurch gekennzeichnet, dass:
    der Überlappungsnenner, m, einen gemeinsamen Faktor mit n1 oder n2 hat.
  6. Verfahren gemäß Anspruch 1 oder 2, wobei der Bereitstellungsschritt ferner gekennzeichnet ist durch:
    Erschaffen eines Vektors der Länge m, der die Längen entweder der überlappenden Teile oder der nicht-überlappenden Teile der Blöcke darstellt, wobei der Durchschnitt der Elemente des Vektors l/m ist, und die Sequenz der Überlappungen sich in dem Vektor zyklisch wiederholt.
  7. Verfahren gemäß Anspruch 1 oder 2, wobei der Ausrichtungsschritt ferner gekennzeichnet ist durch:
    Multiplizieren von unterschiedlichen Blöcken kommendender DFT-Werte mit unterschiedlichen inkrementellen Phasenverschiebungen, die den DFTs der unterschiedlichen Verzögerungen entsprechen.
  8. Verfahren gemäß Anspruch 1 oder 2, wobei der Zeitausrichtungsschritt ferner gekennzeichnet ist durch:
    Multiplizieren der Koeffizienten der Filterantwort in dem Frequenzbereich, H(k), mit der selben inkrementellen Phasenverschiebung.
  9. Sender, der zum Durchführen des Verfahrens von Anspruch 1 oder 2 ausgebildet ist.
  10. Empfänger, der zum Durchführen des Verfahrens von Anspruch 1 oder 2 ausgebildet ist.
EP99970801A 1998-10-16 1999-09-24 Schnelle faltung unter verwendung einer ungeraden tranformation Expired - Lifetime EP1121754B1 (de)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
US09/173,803 US6324559B1 (en) 1998-10-16 1998-10-16 Odd-transform fast convolution
US173803 1998-10-16
PCT/SE1999/001694 WO2000024125A1 (en) 1998-10-16 1999-09-24 Odd-transform fast convolution

Publications (2)

Publication Number Publication Date
EP1121754A1 EP1121754A1 (de) 2001-08-08
EP1121754B1 true EP1121754B1 (de) 2007-07-04

Family

ID=22633563

Family Applications (1)

Application Number Title Priority Date Filing Date
EP99970801A Expired - Lifetime EP1121754B1 (de) 1998-10-16 1999-09-24 Schnelle faltung unter verwendung einer ungeraden tranformation

Country Status (9)

Country Link
US (1) US6324559B1 (de)
EP (1) EP1121754B1 (de)
JP (1) JP4373011B2 (de)
KR (1) KR100655829B1 (de)
CN (1) CN100426672C (de)
AU (1) AU765761B2 (de)
CA (1) CA2346280C (de)
DE (1) DE69936460T2 (de)
WO (1) WO2000024125A1 (de)

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6169789B1 (en) * 1996-12-16 2001-01-02 Sanjay K. Rao Intelligent keyboard system
US6480535B1 (en) * 1999-07-26 2002-11-12 Sony Corporation Generating multi-channel outputs simultaneously in a receiver using polyphase filter
GB2403612B (en) 2000-05-12 2005-02-16 Global Silicon Ltd Radio receiver
FR2817100B1 (fr) * 2000-11-17 2003-08-22 Cit Alcatel Procede pour optimiser les performances d'un emetteur pour systeme de radiocommunications mobiles
GB0116493D0 (en) 2001-07-06 2001-08-29 Koninkl Philips Electronics Nv Receiver having an adaptive filter and method of optimising the filter
DE60210894T2 (de) * 2001-09-25 2006-11-30 Koninklijke Philips Electronics N.V. Filterschaltung für Videodaten im Frequenzbereich
US6792057B2 (en) * 2002-08-29 2004-09-14 Bae Systems Information And Electronic Systems Integration Inc Partial band reconstruction of frequency channelized filters
JP3962785B2 (ja) * 2003-07-02 2007-08-22 テクトロニクス・インターナショナル・セールス・ゲーエムベーハー シグナル・アナライザ及び周波数領域データ生成方法
US7480689B2 (en) * 2004-11-19 2009-01-20 Massachusetts Institute Of Technology Systolic de-multiplexed finite impulse response filter array architecture for linear and non-linear implementations
US8005176B2 (en) * 2007-02-15 2011-08-23 Massachusetts Institute Of Technology Architecture for systolic nonlinear filter processors
JP2011004264A (ja) * 2009-06-19 2011-01-06 Fujitsu Ltd ディジタル信号処理装置およびディジタル信号処理方法
HUE030185T2 (en) 2011-03-28 2017-04-28 Dolby Laboratories Licensing Corp Reduced Transformation for Low Frequency Effect Channel
JP6012066B2 (ja) * 2012-08-01 2016-10-25 Ｎｅｃスペーステクノロジー株式会社 チャネライザ及び信号処理方法
US9286268B2 (en) * 2013-12-12 2016-03-15 Brno University of Technology Method and an apparatus for fast convolution of signals with a one-sided exponential function
US9148162B2 (en) * 2014-01-15 2015-09-29 Guzik Technical Enterprises Digital down converter with equalization
US10044488B1 (en) * 2015-04-21 2018-08-07 Raytheon Company Interpolated channelizer with compensation for non-linear phase offsets
US10211919B2 (en) * 2015-07-09 2019-02-19 Ciena Corporation Mitigation of narrow-band degradations in a broadband communication system
FR3049131B1 (fr) 2016-03-18 2018-04-06 Thales Procede de filtrage d'un signal d'entree numerique et filtre associe
CN107911323A (zh) * 2017-11-03 2018-04-13 西安电子科技大学 基于部分解耦合的联合频相估计方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5270953A (en) * 1991-05-23 1993-12-14 Rockwell International Corporation Fast convolution multiplier
FR2702612B1 (fr) * 1993-03-10 1995-06-09 France Telecom Procede et dispositif de filtrage d'un signal temporel numerique, et application a la correction d'echos dans un canal de transmission .
US5537435A (en) 1994-04-08 1996-07-16 Carney; Ronald Transceiver apparatus employing wideband FFT channelizer with output sample timing adjustment and inverse FFT combiner for multichannel communication network
US6134571A (en) * 1998-04-29 2000-10-17 Hewlett-Packard Company Implicit DST-based filter operating in the DCT domain
SE9802059D0 (sv) 1998-06-10 1998-06-10 Ericsson Telefon Ab L M Digital channeliser and De-shanneliser

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
None *

Also Published As

Publication number Publication date
CA2346280C (en) 2012-04-10
AU765761B2 (en) 2003-09-25
DE69936460D1 (de) 2007-08-16
CN100426672C (zh) 2008-10-15
KR20010080180A (ko) 2001-08-22
WO2000024125A1 (en) 2000-04-27
JP4373011B2 (ja) 2009-11-25
AU1421300A (en) 2000-05-08
CA2346280A1 (en) 2000-04-27
JP2002528947A (ja) 2002-09-03
DE69936460T2 (de) 2008-03-06
EP1121754A1 (de) 2001-08-08
CN1330809A (zh) 2002-01-09
KR100655829B1 (ko) 2006-12-12
US6324559B1 (en) 2001-11-27

Similar Documents

Publication Publication Date Title
EP1114516B1 (de) Flexibilitätsverbesserung eines modifizierten schnellen faltungsalgoritmus
EP1121754B1 (de) Schnelle faltung unter verwendung einer ungeraden tranformation
EP1540941B1 (de) Partialbandrekonstruktion von frequenzkanalisierten filtern
US7227902B2 (en) Method and apparatus for digital channelisation and de-channelisation
US7145972B2 (en) Polyphase channelization system
US7035888B2 (en) Digital sampling rate converter for compensation for drop of in-band signal
US20220200637A1 (en) High-performance filter bank channelizers
EP1114537B1 (de) Veränderter algorithmus zur schnellen faltung niedrigerer komplexität und höherer flexibilität
US6452982B1 (en) Method and system for-down-converting a signal
AU755991B2 (en) Reduced complexity and increased flexibility modified fast convolution algorithm

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20010417

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT SE

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)

RBV Designated contracting states (corrected)

Designated state(s): DE FR GB SE

GRAP Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOSNIGR1

GRAS Grant fee paid

Free format text: ORIGINAL CODE: EPIDOSNIGR3

GRAA (expected) grant

Free format text: ORIGINAL CODE: 0009210

AK Designated contracting states

Kind code of ref document: B1

Designated state(s): DE FR GB SE

REG Reference to a national code

Ref country code: GB

Ref legal event code: FG4D

REF Corresponds to:

Ref document number: 69936460

Country of ref document: DE

Date of ref document: 20070816

Kind code of ref document: P

ET Fr: translation filed
PLBE No opposition filed within time limit

Free format text: ORIGINAL CODE: 0009261

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT

26N No opposition filed

Effective date: 20080407

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: SE

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20071004

REG Reference to a national code

Ref country code: FR

Ref legal event code: PLFP

Year of fee payment: 18

REG Reference to a national code

Ref country code: FR

Ref legal event code: PLFP

Year of fee payment: 19

REG Reference to a national code

Ref country code: FR

Ref legal event code: PLFP

Year of fee payment: 20

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: DE

Payment date: 20180927

Year of fee payment: 20

Ref country code: FR

Payment date: 20180925

Year of fee payment: 20

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: GB

Payment date: 20180927

Year of fee payment: 20

REG Reference to a national code

Ref country code: DE

Ref legal event code: R071

Ref document number: 69936460

Country of ref document: DE

REG Reference to a national code

Ref country code: GB

Ref legal event code: PE20

Expiry date: 20190923

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: GB

Free format text: LAPSE BECAUSE OF EXPIRATION OF PROTECTION

Effective date: 20190923